xref: /linux/fs/ext4/fsync.c (revision e190bfe56841551b1ad5abb42ebd0c4798cc8c01)
1 /*
2  *  linux/fs/ext4/fsync.c
3  *
4  *  Copyright (C) 1993  Stephen Tweedie (sct@redhat.com)
5  *  from
6  *  Copyright (C) 1992  Remy Card (card@masi.ibp.fr)
7  *                      Laboratoire MASI - Institut Blaise Pascal
8  *                      Universite Pierre et Marie Curie (Paris VI)
9  *  from
10  *  linux/fs/minix/truncate.c   Copyright (C) 1991, 1992  Linus Torvalds
11  *
12  *  ext4fs fsync primitive
13  *
14  *  Big-endian to little-endian byte-swapping/bitmaps by
15  *        David S. Miller (davem@caip.rutgers.edu), 1995
16  *
17  *  Removed unnecessary code duplication for little endian machines
18  *  and excessive __inline__s.
19  *        Andi Kleen, 1997
20  *
21  * Major simplications and cleanup - we only need to do the metadata, because
22  * we can depend on generic_block_fdatasync() to sync the data blocks.
23  */
24 
25 #include <linux/time.h>
26 #include <linux/fs.h>
27 #include <linux/sched.h>
28 #include <linux/writeback.h>
29 #include <linux/jbd2.h>
30 #include <linux/blkdev.h>
31 
32 #include "ext4.h"
33 #include "ext4_jbd2.h"
34 
35 #include <trace/events/ext4.h>
36 
37 /*
38  * If we're not journaling and this is a just-created file, we have to
39  * sync our parent directory (if it was freshly created) since
40  * otherwise it will only be written by writeback, leaving a huge
41  * window during which a crash may lose the file.  This may apply for
42  * the parent directory's parent as well, and so on recursively, if
43  * they are also freshly created.
44  */
45 static void ext4_sync_parent(struct inode *inode)
46 {
47 	struct dentry *dentry = NULL;
48 
49 	while (inode && ext4_test_inode_state(inode, EXT4_STATE_NEWENTRY)) {
50 		ext4_clear_inode_state(inode, EXT4_STATE_NEWENTRY);
51 		dentry = list_entry(inode->i_dentry.next,
52 				    struct dentry, d_alias);
53 		if (!dentry || !dentry->d_parent || !dentry->d_parent->d_inode)
54 			break;
55 		inode = dentry->d_parent->d_inode;
56 		sync_mapping_buffers(inode->i_mapping);
57 	}
58 }
59 
60 /*
61  * akpm: A new design for ext4_sync_file().
62  *
63  * This is only called from sys_fsync(), sys_fdatasync() and sys_msync().
64  * There cannot be a transaction open by this task.
65  * Another task could have dirtied this inode.  Its data can be in any
66  * state in the journalling system.
67  *
68  * What we do is just kick off a commit and wait on it.  This will snapshot the
69  * inode to disk.
70  *
71  * i_mutex lock is held when entering and exiting this function
72  */
73 
74 int ext4_sync_file(struct file *file, int datasync)
75 {
76 	struct inode *inode = file->f_mapping->host;
77 	struct ext4_inode_info *ei = EXT4_I(inode);
78 	journal_t *journal = EXT4_SB(inode->i_sb)->s_journal;
79 	int ret;
80 	tid_t commit_tid;
81 
82 	J_ASSERT(ext4_journal_current_handle() == NULL);
83 
84 	trace_ext4_sync_file(file, datasync);
85 
86 	if (inode->i_sb->s_flags & MS_RDONLY)
87 		return 0;
88 
89 	ret = flush_completed_IO(inode);
90 	if (ret < 0)
91 		return ret;
92 
93 	if (!journal) {
94 		ret = generic_file_fsync(file, datasync);
95 		if (!ret && !list_empty(&inode->i_dentry))
96 			ext4_sync_parent(inode);
97 		return ret;
98 	}
99 
100 	/*
101 	 * data=writeback,ordered:
102 	 *  The caller's filemap_fdatawrite()/wait will sync the data.
103 	 *  Metadata is in the journal, we wait for proper transaction to
104 	 *  commit here.
105 	 *
106 	 * data=journal:
107 	 *  filemap_fdatawrite won't do anything (the buffers are clean).
108 	 *  ext4_force_commit will write the file data into the journal and
109 	 *  will wait on that.
110 	 *  filemap_fdatawait() will encounter a ton of newly-dirtied pages
111 	 *  (they were dirtied by commit).  But that's OK - the blocks are
112 	 *  safe in-journal, which is all fsync() needs to ensure.
113 	 */
114 	if (ext4_should_journal_data(inode))
115 		return ext4_force_commit(inode->i_sb);
116 
117 	commit_tid = datasync ? ei->i_datasync_tid : ei->i_sync_tid;
118 	if (jbd2_log_start_commit(journal, commit_tid)) {
119 		/*
120 		 * When the journal is on a different device than the
121 		 * fs data disk, we need to issue the barrier in
122 		 * writeback mode.  (In ordered mode, the jbd2 layer
123 		 * will take care of issuing the barrier.  In
124 		 * data=journal, all of the data blocks are written to
125 		 * the journal device.)
126 		 */
127 		if (ext4_should_writeback_data(inode) &&
128 		    (journal->j_fs_dev != journal->j_dev) &&
129 		    (journal->j_flags & JBD2_BARRIER))
130 			blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL,
131 					NULL, BLKDEV_IFL_WAIT);
132 		ret = jbd2_log_wait_commit(journal, commit_tid);
133 	} else if (journal->j_flags & JBD2_BARRIER)
134 		blkdev_issue_flush(inode->i_sb->s_bdev, GFP_KERNEL, NULL,
135 			BLKDEV_IFL_WAIT);
136 	return ret;
137 }
138